EXPLORE PROCESS BIG DATA Michael Brown … · explore process big data analytics with the uniformance suite michael brown 25 oct 2016

EXPLORE PROCESS BIG DATA

ANALYTICS WITH THE

UNIFORMANCE SUITE

Michael Brown

25 Oct 2016

Honeywell Proprietary - © 2016 by Honeywell International Inc. All rights reserved.

Uniformance® Suite Delivers Digital Intelligence

1

Organize and visualize

data in asset context

Apply powerful analytics

to detect and predict issues

Connect process intelligence

to business KPIs

Capture real-time process

and event data

CollaborationAcross Functions

Management

Maintenance & Reliability

OperationsEngineering

Visualization

Ad-hoc Analysis KPI Dashboard

Notifications


The Industrial Analytics Challenge

2

How can big data and analytics help close this GAP

and deliver real value in the process industries?

40 exabytes (4.0 x 1019) of

unique new information

will be generated worldwide

this year

A major O&G producer has

over 40 Pbyte (1015 or

quadrillion) of data today

and will increase 10X in

coming years

Today

Taking data

to action

Data from sensors

& applications

Analytics GAP


Data in context of the

problem

Closing the Industrial Analytics Gap

3

Open data analytics

platform for people to

collaborate

Simple tools usable by

plant Engineers with

domain knowledge

Fuse sensor data and

application data for

complete understanding


A Pragmatic Definition of Data Analytics

4

Graphic from “Big Data Industry Insights”, Lisa Kart, Gartner webinar, 2015.

• The goal of analytics is to provide information for improved

decisions and actions for economic benefit.

• Note that maximizing automation and minimizing human input are

not always the goals analytics should be suited to the use case.


© 2015 by Honeywell International Inc. All rights reserved.

Uniformance Suite - Run-time Predictive

Analytics Approaches

No Silver Bullet – Hybrid Approaches Needed

Approach Application Technology ComplexityApplic-

ability

1. Physics Model

(1st Principles)

Basic perf mon for broad set of assets

& detection deviation from predicted vs

actual

Embedded in AM & External

via UniSim based on heatLow Many

2. Univariate Prediction

Predicting single variable time to reach

a value – e.g. predict heat exchanger

fouling

Regression w error correction

(H_TimeFit)

3. Adaptive Filtering/

Thresholding

Anomaly detection for Equipment

(temp, press, vib)

Data cleansing & compare

current to historical averages

4. Multivariate Pattern

Detect behavior of group of sensors

according to learned/historical

expectations

Detected patterns (equations)

with rules detecting abnormal

relationships

5. Multivariate Early

Event Detection

Broad set of process and equipment

monitoring scenarios

Statistical pattern detection

and recognition including OLS,

PLS, PCA, Neural Nets, etc.

(Honeywell & 3rd Party)

6. Machine Learning

Capture insights from Big Data to build

better algorithms (e.g. Aero APU

example)

Big Data using variety of data

sources including maintenance

and reliability data

High Few


6


7

A

Consolidated and contextualized data

Capture | Uniformance® PHD

8

• Scalable - from site to enterprise to cloud

• Data Fusion: More than a process

historian – includes event and

application data

• Built-in engineering knowledge

• Built-in calculation and engineering unit

conversion capability

• Use with asset model for enterprise

integration

• Visual process analytics search tools

Capture

Consolidate

Contextualize

$

fx

A


9

B

Flexible web-enabled AGILE collaboration

Visualize | Uniformance® Insight

10

• Powerful ad-hoc trends and

workspaces

• True thin-client application

• Available on a wide variety of devices

• Collaborative interaction & sharing

• Powerful visual analytics search engine

• Connect to process historians,

application data & 3rd party data

B


11

C


12

• Enables Advanced Analysis

• Extreme scalability, Scalable on the fly

• Millions of tags, Millions of events per

second

• Advanced aggregation and calculations

• Advanced data synchronization

• Enterprise data fusion

• Support for enterprise security

Experion PKS

Host Server

Experion PKS

Cold Stand-by

UniformancePHD Shadow and CollectorsSite A

Experion PKS

Host Server

Experion PKS

Cold Stand-by

UniformancePHD Shadow and Collectors

Site B

Honeywell Cloud Historian

OPC UA & Third Party Historians

Enterprise | Cloud HistorianC


13

D


Visual Analytics for Equipment Monitoring

• Designing monitoring rules can be time consuming as it is an

iterative approach of exploring anomaly behavior and testing rule

designs against historical data with subsequent iterations

• Honeywell is pursuing a Visual Data Analytics approach that

provides visual data exploration and visualization techniques to

augment the current workflow and enable SMEs to develop

analytics in an agile and iterative manner

(Forensics) Has this timeseries

combination happened before? Where

and when? What else happened before

and after it?

(Design) I am designing a new

monitoring rule. I want to find all

examples in my data where temperature

was rising while load was staying roughly

constant.

(Exploration) I want to find all examples

where something interesting happened

on compressor 7 between July and

October 2012.

?

(Rule discovery) I want to search for all

examples where a outage and workorder

were not preceded by an Alert. Then

decide if I need a new rule.

outage

workorder

Example Queries

14


• Data Fusion

- Historians, Maintenance, LIMS,

Financial, …

- All meta data is indexed

- On-line, no data lake

• Capsule Series

- “Time periods of interest”

- Simultaneously time series and

transactional

• Scalability

- Users’ formulas automatically

scale

• Searching

- Patterns

- Limits / Boundaries

- Logical

• Interactive

- On-the-fly calcs

- Collaborative UI

• Historical Benchmarking *

- Mode or conditioned based calcs

Seeq Corporation Proprietary - Limited Distribution 15

* Patent Pending

Saves 95% of the effort required by engineers to

explore their ideas for surveillance, troubleshooting,

and improvements => data driven decisions

Seeq | ”Workbench” Innovations D15


See the Invisible

Supply Chain & Trading Data

Lab/Quality Data Inventory DataMaintenance Data HSE Data

Planning & Scheduling Data

Reliability Data

Operations & Personnel Data

Financial and Business Data

Process and Control Alarm &

Event data

Time-in-State “Capsules” add context to time series data for exploratory visual analysis:

Shift B

Sensor Data

Pressure above 300 psi for 30 min

Shutdown for Maintenance

Historian

16


Visual Analytics Demonstration17


18

E

Moving to Early Event Detection

Predict | Uniformance® Asset Sentinel

19

• Powerful analytics with asset-based calculations and fault models

• Integrated workflow from alert notification, troubleshooting, to problem resolution

• Embedded process content (pump,

compressors, turbines, etc.)

• Email & mobile notifications for faster

response

• Integrated visual analytics workflow

Embedded & User Defined

Models

Calculations & Fault

Models

Make Problems Visible

Dashboard

s

Notifications

Continuous

Monitoring

Heat Exchanger

Gas / SteamTurbine

Compressor

Furnace

Asset Model

Trends /

Displays

C


© 2015 by Honeywell International Inc. All rights reserved.

Visual Analytics: Workflow with Domain Expertise

Scalable workflow delivering analytics environment with 90% time savings

20

Anomaly

Definition

Failure

Examples

8 candidate post lube failures found in data

Signal

Selection

Use domain expertise to select

shortlist of tags relevant for fault

detection

Review

FailuresRapidly query system behavior

for selected failure events

Design RuleDesign rule based from observed

failure events & implement in

Sentinel

Test Rule Test rule against historical data

Deploy Rule

Desig

n I

tera

tion

Ite

rate

on

ru

le d

esig

n/t

un

ing

to

ach

ieve

de

sir

ed

ba

lan

ce

of tr

ue

po

sitiv

e v

s.

fals

e p

ositiv

es

Alert indication of

anomaly/exception from process

Rapidly determine failure

occurrences in historical data

Real-time analytics deployed to

online runtime monitoring system

Post Lube Alert After compressor shutdown lube oil pressure must remain higher than 10 psi for 20-30 minutes or until oil temperature is low.

Out of 8 candidates, 4 were true post lube failures

True Failure Not a failure

Positive detection of all 4 post

lube failures with no false

positives. Rule validated on

other platforms.

Rule deployed in Asset

Sentinel for real-time

detection

Maintain RuleDeploy rule to online runtime

monitoring system

Insig

ht

Vis

ua

l A

naly

tic

s

Vis

ua

l

An

aly

tic

s As

se

t S

en

tin

el


Uniformance Analytics Suite Overview

Run-time Analytics

Unit / Site

Process Data

Real-time & Historical

(Small Data)

Normal & Abnormal

• First Principals

• Statistical

• State estimation

Model

Event

Detection Deviation Detection

• Heuristic

• Trained

Off-Line Analytics

Unit / Site / Multi-Site

Visual Data Analytics

• Pattern search

• Value Search

• Combinations

• Cleanse / Filter

Process

Engineer

Data

Scientist

Skillsets Required

Statistical Analytics

• Multivariate

statistical (PCA,

PLS…)

Machine Learning

Big Data

• Data Vol. & variety

(unstructured / text)

• Volume of data

• Feature Extraction

• ML (Random Forest,

SVM, Naïve Bays…)

Additional

Models / Rules

Where is sweet spot

for value generation

and scalability?

21


22


23

EXPLORE PROCESS BIG DATA Michael Brown … · explore process big data analytics with the uniformance suite michael brown 25 oct 2016

Documents